I’ve spoken previously about the lack of moat of wrapper-AI startups. And the early startups from 2023 had this problem. They were just a method to obfuscate the prompt. And “prompt is our IP” has never been a moat.

However, I’ve been hearing pitches with dozens of AI startups over the last few months, and the product, UI and distribution moats are legit. Founders have finally begun to create AI-native SaaS that uses 3rd party LLMs, a mix of original and synthetic training data, as well as some legit IP.

It’s the age-old argument of product-manufacturing or product-distribution.

Sounds like, today we are in the distribution-is-king phase of AI.

These startups can switch the underlying LLM’s at will, but as long as they solve for legit customer pain points, their distribution and product moats will allow them to retain customers.

“LLM is a commodity for a price” – is this what is spooked everyone in the markets?

I still think the market reaction to DeepSeek is irrational and over-done. I mean, it’s not like China has found a replacement to Nvidia GPUs (yet).

Comments

comments